An Efficient Multiscale Pyramid Attention Network for Face Detection in Surveillance Images

Table 1

The architecture of backbone.

Stage i	Operator	Resolution	Channels	Layers
1	Conv3 × 3	512 × 512	45	1
2	MBConv1	256 × 256	22	1
3	MBConv6	256 × 256	24	4
4	MBConv6	128 × 128	40	4
5	MBConv6	128 × 128	80	5
6	MBConv6	64 × 64	112	5
7	MBConv6	64 × 64	192	7
8	MBConv6	32 × 32	320	2
9	Conv1 × 1 and pooling and FC	32 × 32	1280	1

MBConv denotes mobile inverted convolutional bottleneck.